On Tue, Oct 4, 2011 at 10:04 AM, George Tzotzos <gtzotzos.me.com> wrote:
> Just a small clarification.
>
> The part "In this case, you specified 5 to 45 via increments of 5 (9 frames
> total)" is well understood.
>
> However, I was referring to a separate run for which I set the
> nmstartframe=5 and the nmendframe=200 with nminterval=10.
>
> I was expecting that the calculation would be performed on 40 frames
> instead of the 5 that is reported. This is what is confusing me.
>
Ah. Are you using 8 processors? That would explain why the master is only
computing 5 (9 processors would give you the same result). The number of
frames that the master is computing is *not* the total number of frames
(unless you're only using 1 processor in the MPI version).
> A second broader question is whether it is legitimate to draw the first 50
> frames for each of the 5 trajectories for the calculation. Doesn't this give
> biased sampling?
>
Depends on your trajectories. If consecutive snapshots are correlated, then
certainly. If not, it's probably OK.
HTH,
Jason
>
> Thanks for the patience and explanations
>
> George
>
> On Oct 4, 2011, at 3:47 PM, Jason Swails wrote:
>
> > On Tue, Oct 4, 2011 at 9:35 AM, George Tzotzos <gtzotzos.me.com> wrote:
> >
> >> Jason,
> >>
> >> Thanks for the explanation. There's a second point that I'd like to
> clarify
> >> - if possible.
> >>
> >> My 10ns trajectory is made up of 5 individual trajectories of 2ns and
> 200
> >> frames each.
> >>
> >> If I use an input file such as the one below
> >>
> >> &general
> >> endframe=50, keep_files=2,
> >> /
> >> &nmode
> >> nmstartframe=5, nmendframe=45,
> >> nminterval=5, nmode_igb=1, nmode_istrng=0.1,
> >> /
> >>
> >> I get
> >>
> >> Calculations performed using 250 frames.
> >> |NMODE calculations performed using 9 frames.
> >>
> >> Setting the endframe=200 and
> >> nmstartframe=5, nmendframe=200,
> >> nminterval=10
> >>
> >> I get "Master thread is calculating normal modes for 5 frames".
> >>
> >> It's a bit esoteric for me the way the program works out the number of
> >> frames. Any explanation would be most welcome.
> >>
> >
> > I agree. It applies "startframe, endframe, interval" to all of the
> > trajectories for the solvation free energy terms. That is, if you have 5
> > total trajectories, then your input file will draw the first 50 frames of
> > each one (for 250 total). Note that this will change with the new
> version
> > to behave more like you expect (it will draw the first 50 frames of the
> sum
> > of all trajectories, leading to 50 frames total).
> >
> > For normal mode calculations, the nmstartframe, nmendframe, and
> nminterval
> > are applied to the _MMPBSA_complex.mdcrd (or _MMPBSA_complex.nc) file,
> since
> > it's assumed you'd only ever want to do nmode calculations on a subset of
> > the frames you do all other calculation types on. In this case, you
> > specified 5 to 45 via increments of 5 (9 frames total).
> >
> > Afterwards, it says that the master thread is calculating normal modes
> for 5
> > frames, which tells me that you're using 2 processors. In this case, it
> > splits up the frames as evenly as possible, but you can't divide 9 frames
> > evenly between 2 processors, so the master gets 5 and the other one gets
> 4.
> >
> > HTH,
> > Jason
> >
> >
> >> Regards
> >>
> >> George
> >>
> >>
> >> On Oct 3, 2011, at 9:47 PM, Jason Swails wrote:
> >>
> >>> On Mon, Oct 3, 2011 at 2:01 PM, George Tzotzos <gtzotzos.me.com>
> wrote:
> >>>
> >>>> I'm running MMPBSA.MPI on 12 processors. I'm trying to compute the
> >> entropy
> >>>> (nmode) of a protein/ligand complex from a 10ns trajectory (1000)
> >> frames.
> >>>>
> >>>> My entropy.in is:
> >>>>
> >>>> Input file for running entropy calculations using NMode
> >>>> &general
> >>>> endframe=1000, keep_files=2,
> >>>> /
> >>>> &nmode
> >>>> nmstartframe=10, nmendframe=990,
> >>>> nminterval=10, nmode_igb=1, nmode_istrng=0.1,
> >>>> /
> >>>>
> >>>>
> >>>> My estimate is that it'll take at least 10 hours to get results.
> >>>>
> >>>
> >>> Normal mode estimates are fairly unreliable. As expensive as the
> Hessian
> >>> construction/diagonalization is (we're helped by the fact that it's an
> >>> analytical Hessian, I think), in my experience the minimization takes
> >>> significantly longer. However, the number of steps required to
> minimize
> >>> within tolerable limits obviously depends on how close the structure is
> >> to a
> >>> minimum initially. MMPBSA.MPI's parallelization scheme is naive,
> though,
> >>> and assumes each will take the same amount of time, which hurts
> parallel
> >>> scaling for normal mode calcs in most cases.
> >>>
> >>>
> >>>>
> >>>> My question is whether increasing nminterval to 100 would compensate
> the
> >>>> accuracy of the calculation.
> >>>>
> >>>
> >>> Unlikely (though maybe increase it to 50 instead). Keep in mind that
> >> NMODE
> >>> frames are minimized, so in all likelihood many of your snapshots are
> >> very
> >>> structurally similar to others (if they reduce to the same local
> >> minimum).
> >>> A better approach may be to perform normal mode analysis on clustered
> >>> trajectories, and only take one snapshot from each cluster (or
> something
> >>> like that). You can always check existing literature to see what
> they've
> >>> done as well.
> >>>
> >>> HTH,
> >>> Jason
> >>>
> >>>
> >>>> Thanks in advance
> >>>>
> >>>> George
> >>>>
> >>>>
> >>>>
> >>>>
> >>>> _______________________________________________
> >>>> AMBER mailing list
> >>>> AMBER.ambermd.org
> >>>> http://lists.ambermd.org/mailman/listinfo/amber
> >>>>
> >>>
> >>>
> >>>
> >>> --
> >>> Jason M. Swails
> >>> Quantum Theory Project,
> >>> University of Florida
> >>> Ph.D. Candidate
> >>> 352-392-4032
> >>> _______________________________________________
> >>> AMBER mailing list
> >>> AMBER.ambermd.org
> >>> http://lists.ambermd.org/mailman/listinfo/amber
> >>
> >>
> >> _______________________________________________
> >> AMBER mailing list
> >> AMBER.ambermd.org
> >> http://lists.ambermd.org/mailman/listinfo/amber
> >>
> >
> >
> >
> > --
> > Jason M. Swails
> > Quantum Theory Project,
> > University of Florida
> > Ph.D. Candidate
> > 352-392-4032
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Oct 04 2011 - 08:30:02 PDT
